Language Service Parser and Scanner (Managed Package Framework)

The parser is the heart of the language service. The Managed Package Framework (MPF) language classes require a language parser to select information about the code being displayed. A parser separates the text into lexical tokens and then identifies those tokens by type and functionality.

The following is a C# method.

namespace MyNamespace
    class MyClass
        public void MyFunction(int arg1)
            int var1 = arg1;

In this example, the tokens are the words and punctuation marks. The kinds of tokens are as follows.

Token Name

Token Type

namespace, class, public, void, int




{ } ( ) ;


MyNamespace, MyClass, MyFunction, arg1, var1











local variable

The role of the parser is to identify the tokens. Some tokens can have more than one type. After the parser has identified the tokens, the language service can use the information to provide helpful features, such as syntax highlighting, brace matching, and the IntelliSense operations.

A language service parser is not the same as a parser used as part of a compiler. However, this kind of parser needs to use both a scanner and a parser, in the same way as a compiler parser.

  • A scanner is used to identify types of tokens. This information is used for syntax highlighting and for quickly identifying token types that can trigger other operations, for example, brace matching. This scanner is represented by the IScanner interface.

  • A parser is used to describe the functions and scope of the tokens. This information is used in IntelliSense operations to identify language elements, such as methods, variables, parameters, and declarations, and to provide lists of members and method signatures based on context. This parser is also used to locate matching language element pairs, such as braces and parentheses. This parser is accessed through the ParseSource method in the LanguageService class.

How you implement a scanner and parser for your language service is up to you. Several resources are available that describe how parsers work and how to write your own parser. Also, several free and commercial products are available that help in creating a parser.

Unlike a parser that is used as part of a compiler (where the tokens are converted to some form of executable code), a language service parser can be called for many different reasons and in many different contexts. How you implement this approach in the ParseSource method in the LanguageService class is up to you. It is important to keep in mind that the ParseSource method might be called on a background thread.

Caution note Caution

The ParseRequest structure contains a reference to the IVsTextView object. This IVsTextView object cannot be used in the background thread. In fact, many of the base MPF classes cannot be used in the background thread. These include the Source, ViewFilter, CodeWindowManager classes, and any other class that directly or indirectly communicates with the view.

This parser typically parses the whole source file the first time it is called or when the parse reason value of Check is given. Subsequent calls to the ParseSource method handle a small part of the parsed code and can be executed much more quickly by using the results of the previous full parse operation. The ParseSource method communicates the results of the parsing operation through the AuthoringSink and AuthoringScope objects. The AuthoringSink object is used to collect information for a specific parsing reason, for example, information about the spans of matching braces or method signatures that have parameter lists. The AuthoringScope provides collections of declarations and method signatures and also support for the Go To advanced edit option (Go to Definition, Go to Declaration, Go to Reference).

You must also implement a scanner that implements IScanner. However, because this scanner operates on a line-by-line basis through the Colorizer class, it is typically easier to implement. At the beginning of each line, the MPF gives the Colorizer class a value to use as a state variable that is passed to the scanner. At the end of each line, the scanner returns the updated state variable. The MPF caches this state information for each line so that the scanner can start parsing from any line without having to start at the beginning of the source file. This fast scanning of a single line allows the editor to provide fast feedback to the user.

This example shows the flow of control for matching a closing brace that the user has typed. In this process, the scanner that is used for colorization is also used to determine the type of token and whether the token can trigger a match-brace operation. If the trigger is found, the ParseSource method is called to find the matching brace. Finally, the two braces are highlighted.

Even though braces are used in the names of triggers and parse reasons, this process is not limited to actual braces. Any pair of characters that that is specified to be a matching pair is supported. Examples include ( and ), < and >, and [ and ].

Assume that the language service supports matching braces.

  1. The user types a closing curly brace (}).

  2. The curly brace is inserted at the cursor in the source file and the cursor is advanced by one.

  3. The OnCommand method in the Source class is called with the typed closing brace.

  4. The OnCommand method calls the GetTokenInfo method in the Source class to obtain the token at the position just before the current cursor position. This token corresponds to the typed closing brace).

    1. The GetTokenInfo method calls the GetLineInfo method on the Colorizer object to obtain all tokens on the current line.

    2. The GetLineInfo method calls the SetSource method on the IScanner object with the text of the current line.

    3. The GetLineInfo method repeatedly calls the ScanTokenAndProvideInfoAboutIt method on the IScanner object to gather all tokens from the current line.

    4. The GetTokenInfo method calls a private method in the Source class to obtain the token that contains the desired position, and passes in the list of tokens obtained from the GetLineInfo method.

  5. The OnCommand method looks for a token trigger flag of MatchBraces on the token that is returned from the GetTokenInfo method; that is, the token that represents the closing brace).

  6. If the trigger flag of MatchBraces is found, the MatchBraces method in the Source class is called.

  7. The MatchBraces method starts a parsing operation with the parse reason value of HighlightBraces. This operation ultimately calls the ParseSource method on the LanguageService class. If asynchronous parsing is enabled, this call to the ParseSource method occurs on a background thread.

  8. When the parsing operation is finished, an internal completion handler (also known as a callback method) named HandleMatchBracesResponse is called in the Source class. This call is made automatically by the LanguageService base class, not by the parser.

  9. The HandleMatchBracesResponse method obtains a list of spans from the AuthoringSink object that is stored in the ParseRequest object. (A span is a TextSpan structure that specifies a range of lines and characters in the source file.) This list of spans typically contains two spans, one each for the opening and closing braces.

  10. The HandleBracesResponse method calls the HighlightMatchingBrace method on the IVsTextView object that is stored in the ParseRequest object. This highlights the given spans.

  11. If the LanguagePreferences property EnableShowMatchingBrace is enabled, the HandleBracesResponse method obtains the text that is encompassed by the matching span and displays the first 80 characters of that span in the status bar. This works best if the ParseSource method includes the language element that accompanies the matching pair. For more information, see the EnableShowMatchingBrace property.

  12. Done.

The matching braces operation is typically limited to simple pairs of language elements. More complex elements, such as matching triples ("if(…)", "{" and "}", or "else", "{" and "}"), can be highlighted as part of a word-completion operation. For example, when the "else" word is finished, the matching "if" statement can be highlighted. If there were a series of if/else if statements, all of them could be highlighted by using the same mechanism as matching braces. The Source base class already supports this, as follows: The scanner must return the token trigger value MatchBraces combined with the trigger value MemberSelect for the token that is before the cursor position.

For more information, see Brace Matching (Managed Package Framework).

Colorizing source code is straightforward, just identify the type of token and return color information about that type. The Colorizer class acts as the intermediary between the editor and the scanner to provide color information about every token. The Colorizer class uses the IScanner object to help in colorizing a line and also to gather state information for all lines in the source file. In the MPF language service classes, the Colorizer class does not have to be overridden because it communicates with the scanner only through the IScanner interface. You supply the object that implements the IScanner interface by overriding the GetScanner method on the LanguageService class.

The IScanner scanner is given a line of source code through the SetSource method. Calls to the ScanTokenAndProvideInfoAboutIt method are repeated to obtain the next token in the line until the line is exhausted of tokens. For colorization, the MPF treats all source code as a sequence of lines. Therefore, the scanner must be able to cope with source coming at it as lines. In addition, any line can be passed to the scanner at any time, and the only guarantee is that the scanner receives the state variable from the line before the line about to be scanned.

The Colorizer class is also used to identify token triggers. These triggers tell the MPF that a particular token can initiate a more complex operation, such as word completion or matching braces. Because identifying such triggers must be fast and must occur at any location, the scanner is best suited for this task.

For more information, see Syntax Colorizing (Managed Package Framework).

Parsing for functionality and scope requires more effort than just identifying the types of tokens that are encountered. The parser has to identify not only the type of token, but also the functionality for which the token is used. For example, an identifier is just a name, but in your language, an identifier could be the name of a class, namespace, method, or variable, depending on the context. The general type of the token may be an identifier, but the identifier may also have other meanings, depending on what it is and where it is defined. This identification requires the parser to have more extensive knowledge about the language that is being parsed. This is where the AuthoringSink class comes in. The AuthoringSink class collects information about identifiers, methods, matching language pairs (such as braces and parentheses), and language triples (similar to language pairs except that there are three parts, for example, "foreach()" "{" and "}"). In addition, you can override the AuthoringSink class to support code identification, which is used in early validation of breakpoints so that the debugger does not have to be loaded, and the Autos debugging window, which shows local variables and parameters automatically when a program is being debugged and requires the parser to identify appropriate local variables and parameters in addition to those that the debugger presents.

The AuthoringSink object is passed to the parser as part of the ParseRequest object, and a new AuthoringSink object is created every time that a new ParseRequest object is created. In addition, the ParseSource method must return an AuthoringScope object, which is used to handle various IntelliSense operations. The AuthoringScope object maintains a list for declarations and a list for methods, either of which is populated, depending on the reason for parsing. The AuthoringScope class must be implemented.